이미지 텍스트 확인
Check for updates
OPEN
Predicting sexfrom retinal fundus
photographs using automated
deeplearning
Edward Korot! , Nikolas Pontikos’ , Xiaoxuan Liu1,2,? Siegfried K Wagner’ Livia Faes1 ”
Josef Huemer1s, Konstantinos Balaskas? Alastair K. Denniston12,3,6 Anthony Khawajai프 &
PearseA Keaneiz
Deeplearning may transform health care but model development has largely been dependent on
availability Of advanced technical expertise Herein we present the development ofa deeplearning
model by clinicians without codingr which predicts reported sexfrom retinal fundus photographs
model was trained on 84,743 retinalfundus photosfrom the UK Biobank dataset. External validation
was
performed on 252 fundus photosfroma tertiary ophthalmic referral center Forinternal
validation, the area under the receiver operating characteristic curve (AUROC) ofthe codefree deep
learning (CFDL) model was 0.93.Sensitivity, specificity, positive predictive value (PPV) and accuracy
(ACC) were 88.896, 83.69, 87.39 and 86.59, andfor externalvalidation were 83.99, 72.29, 78.296
and 78.69 respectively Clinicians are currently Unaware Of distinct retinal feature variations between
males and females, highlighting the importance of model explainabilityfor this task The model
performed significantly worse whenfoveal pathology was presentinthe external validation dataset,
ACC: 69.4%, compared to 85.4%in healthy eyes, suggesting thefoveaisa salient region for model
performance OR (959CI): 0.36 (0.19,0.70) p =0.0022. Automated machine learning (AutoML) may
enable clinician-driven automated discovery Of novelinsights and disease biomarkers.
눈동자만 보고 70~90% 확률로 남자인지 여자인지 구분 가능
사람은 전문의조차도 못 맞춘다고 함.
